#Http benchmarks
This is a brief explanation on how I setup various webservers to compare with uThreads webserver. Note that this benchmark is testing the how many requests/second webserver can handle, thus we are not going to read data from disk, so all requests are server from the memory for all web servers.
##wrk
First, you need to install wrk which is a http benchmarking tool. All benchmarks are run with
wrk -d 15s -t $T -c 200 http://server:port
where $T is the number of threads and server and port is determined from the webserver. Usually running wrk with the same number of worker threads as the webserver is enough. However, if the CPU utilization of wrk is 100% while the CPU utilization of webserver threads are less that 100%, the number of wrk threads should be increased. The goal of the benchmark is to push the webserver to the max CPU limit.
To run uThreads webserver, simply compile and install uThreads and do a make test and run the webserver from $ROOT_DRIECTORY/bin/webserver. You can modify the number of threads (kernel threads) by passing it as the first argument.
./bin/webserver $T
where $T is the number of threads. Note, that the number of threads here always mean number of worker threads + 1, which means if you pass 4, the number of worker threads will be 3 and there will be 1 poller thread. For small number of threads (less than 8) it is safe to ignore the poller thread and if you want to test the system with 3 threads e.g., pass 4 to run it with 4 worker threads. To be more fair, you can always bind the threads to fix number of cores for all the benchmarks.
You can find the nodejs code that I used under nodejs directory, simply run node server.js. Since a single nodejs process is single threaded, I do not run experiments with more number of threads.
You can find the fasthttp code under fasthttp. First, build the code go build server.go and then run it using
GOMAXPROCS=$T ./server
where $T is the number of threads.
You can download it from here, and also download libhoard from here. libhoard is a more efficient memory allocater said to improve the performance of cppsp. After compiling both, under www folder in cppsp, create a file called hello.cppsp and put Hello Wrold! in the file. Then, issue the following:
LD_PRELOAD=/path_to_libhoard/libhoard.so ./run_example -t $T
where $T determines the number of threads. You can access the server through http://localhost:16969/hello.cppsp. Make sure you call the URL once with your webbrowser, as cppsp loads the content of the file into cache and the very first request is slow to server. After that, the file content is being served from the cache in memory.